Double-threshold validation tool (DTVT): From landslide hazard maps to operational early warning systems

Landslide early warning systems help prevent loss of life and economic damage by providing timely warnings. Typically based on rainfall thresholds, these systems often lack reliability as they incorporate only rainfall as input data. Machine learning offers a solution by incorporating multiple geoenvironmental and climatic factors, and recent advancements have enabled the inclusion of both spatial and temporal dimensions, producing accurate pixel-based landslide hazard maps (LHMs). However, the implementation of these maps in warning systems faces three main challenges: the lack of a standardized validation, the mismatch between the pixeled resolution and the larger spatial units needed for territorial early warning, and the difficulty in establishing a consistent criterion for issuing increasing levels of alert. To address these issues, this study presents an innovative approach based on the Double-Threshold Validation Tool (DTVT); a tool designed to standardize the validation of landslide hazard maps by aggregating pixels into wider spatial units through a “failure probability-instability diffusion” pair of thresholds. The tool was applied to the 16–17 May 2023 landslide event in Florence (Italy) to demonstrate its functioning. This application successfully translated complex pixel-based predictions into a multi-level warning system, bridging the gap between predictive modelling and emergency response.